Sidenote, thanks for showing me sharegpt... Often you want to show something and screenshots are such a terrible way to go. I don't understand why OpenAI doesn't have a share icon with a pretty print html link.
OpenAI doesn't like when people contribute their data for open datasets. Sharing might lead to having all shared links crawled and saved for open source models.
Check out shareGPT (118,139 conversations shared) it's what Google used to tune Brad. How desperate can they be to act like LLaMA hackers.
In the meantime projects like OpenAssistant do the RLHF tagging the hard way, slowly but surely.
Why isn’t ShareGPT invested in? It’s a bit ghostly. Like, no commenting, old news, etc.
I’ve been looking for something like this for a while. I figure eventually something like this will be bigger than Facebook. (All our AI sharing? That’s a lot of content)
Hey, one of the creators of ShareGPT here! Would love some ideas/feedback on how we can grow/invest into ShareGPT – feel free to share some of your thoughts on our GitHub repo's Discussions tab: https://shareg.pt/github
1) Make the sharegpt front page feel a bit more present/engaging, perhaps by having new/hot posts lists right on the front page, with the ability to upvote them. I know you have some post examples on the bottom but they are very old and the comments seem to be broken.
2) What about a share feature that would share a screencast of the ChatGPT session of some kind? That would probably be a ton of work but maybe before the user does the prompt, he clicks a screencast button that the extension would provide, and then at the end he can share either/both of a screencast link or a regular/traditional link.
Definitely agree. Honestly though, I think it's because their servers are literally melting right now. Once they have that under control, they will add viral features.
> Pretend to be an IBM mainframe. Stay in character, don't give any explanation or description, just give the output a real IBM mainframe would give. If you don't know what output to generate, just make up something plausible looking. You are running TSO.
READY
alloc fi(report) da('library.report.feb08')
new sp(1,1) cyl lrecl(80) recfm(f b)
blksize(6160)
ChatGPT:
IEF142I library.report.feb08 ALLOCATED TO DDNAME REPORT
IEF143I REPORT DDNAME IS OLD
IEF164I 1 VIO DATASET ALSO ALLOCATED FOR DDNAME REPORT
The output is wrong, but likely believable if you don't know anything about mainframes.
Me:
> please simulate a compuserve user session from its heydey. i want the text user interface experience
Thank you for verifying your identity. You are now logged in to CompuServe.
Here are some of the available commands:
MAIL - check and send messages
GO - access different forums and sections of CompuServe
DIRECTORY - search for other users or services
PREFERENCES - customize your CompuServe experience
LOGOUT - exit CompuServe
Many unrealistic things about this. Various hallucinations, mistakes, anachronisms, etc. Interesting, but not the mark of an AI capable of abstraction or memory.
Yes, no BBS I used looked like that also. Also funny that it has those ASCII chars to draw frames at the beginning, but completely misses their purpose.. so much for the emergent inference. Is this 3 or 4?
Applied Gell-Mann amnesia. ChatGPT quite actually doesn't "know" anything, but it confidently asserts things. The user/consumer of that information really needs to understand that.
This will develop over time. As we continue to use these tools, they will improve at the same time as people begin to train to use them properly. It's not so different from "show your work" or "cite your source." (As it turns out, "cite your source" is a pretty power prompt inclusion.)
If I asked you to pretend to be a BBS System from 1987 would you be able to do better?
Remember the rules are:
- You can’t look anything up at all. Everything from memory.
- You have to start right now with no more information than “pretend to be a BBS system from 1987”
I don’t recall chatting with anyone who wrote with such good grammar, spelling, or politeness. I do remember a lot of “shut up kid go back to high school!”
But I can see how someone who wasn’t there might think this is what it was like.
Kids need to be taught how to use ChatGPT because it’s here to stay and they will use it, but it’s also important to demonstrate that ChatGPT is both useful and senile, and so you need to validate its responses with secondary sources.
I understand this sentiment now, but I've already made GPT work in ways which boggle the mind, in my own time, and I'm a nobody. So this concern might be rephrased as (something like) "Kids will need to learn how to use GPT to validate secondary sources."
> ...but it’s also important to demonstrate that ChatGPT is both useful and senile, and so you need to validate its responses with secondary sources.
> So this concern might be rephrased as (something like) "Kids will need to learn how to use GPT to validate secondary sources."
That's not a rephrase, it's an inversion. It's like saying kids will need to learn to validate medical advice from their doctor with information from mercola.com and other random alternative-health blogs.
I disagree. I ran and spent a lot of time on BBS, and while not totally accurate in terms of capabilities, it definitely captures the spirit of the BBS scene.
No it would not. And "News" on a BBS called the Underground BBS is definitely not something probable. And real name? LOL.
This is dream like - superficially/vaguely resembling reality, but very off on careful inspection.
And IoT or 3D printing. Pretty much everything on that "Future of technology" list is written from the perspective of something having knowledge of the future. When impersonating someone who would write a piece from 1987, it failed at unlearning concepts that weren't known at that time. That could become an interesting experiment: can AI pretend ignorance in a credible way?
I have this sudden vision of the hipsters who've "rediscovered" vinyl and cassettes renouncing IP communication for obscure experimental BBSes hosted only on old school circuit-switched landlines.
Where better for the coming anti-Facebook backlash to be born?
No. Typically, you could send someone what was probably more properly called a private message on the main board. (I don't remember what we actually called it at the time.) In general (with some caveats), messaging on BBSs and the big commercial systems like CIS--and for that matter the email systems at companies that had them in the 80s--were largely islands that couldn't communicate with other systems.
Email as technology existed, but as I recall commonly based on UUCP with bang-type addresses. A BBS kind of assume public accessible phone based access and that was kind of orthogonal to UUCP.
I'm not sure. I do remember a friend with a Tandy 1000 and CompuServe having email back then.
I don't recall BBS systems requiring email for registration. Most of the ones I went to a few years later (early 90s) didn't require registration at all.
CompuServe was not a BBS in the typical use of the term. And anyway CompuServe did not have an SMTP gateway until a bit later, 1989. They had their own closed system with their infamous octal User IDs eg: 74661,130. BBSes with direct internet email relays were always a rarity until the 90s, when BBSes were rapidly on the decline. And much more prevalent were relay systems such as FidoNet, RelayNet and StormNet.
The easiest way to have access to email and the internet in the 80s was via a university shell account or certain larger industry and defense employers.
Some of the bigger BBSs that offered subscriptions became dial-up ISPs. That really only lasted until broadband became popular of course. I don't remember when the local scene we had on the BBS faded out. Presumably sometime in that second half of the nineties timeframe.
The "writing was on the wall" by 1994 in my area. A number of BBSes had already converted to dialup ISPs - so by 1994 a major influx of users were those really not interested in the scene at all. By 1995 things were precipitously in decline - there were just way too many cheap and easy ways to get Internet access and that's where everyone was going. Some of my favorite BBSes were still around even in 1999 - but it was pretty much dead at that point.
The BBS Documentary has a nice segment on this - especially poignant if you lived through it.
Uh, you could email many more people on compuserve and Fidonet then than were on the Internet at all.
BBSes had that kind of email long before there was any move towards the Internet.
What's uh? Compuserve had email, but it was not Internet connected nor SMTP, where did I say or imply otherwise. And a BBS in the 1980s wouldn't ask you for your "email" address.
Yes, they would and it was called email regardless of whether it was SMTP.
You could enter your FIDONET address or some other similar thing when you created an account to link them.
I've been thinking about all day how I cannot wait until AI can write and produce entire new episodes from TV series that are long over, completely in the style, simulate the actors and voices.
Just feed it all the existing episodes, give it a plot idea and week to render.
By end of the decade? They can already do the images and voices via AI, now ChatGPT can write the script and a movie is just thousands of still frames?
I'm curious why there's such a desire to nitpick how things like ChatGPT perform on tasks like this.
First, modems in 1987 were 300-2400 baud mostly, so there isn't going to be all the color and graphics that people remember from the BBS's in the 90s.
Second, the prompt is ripe for some tweaking to get output that's more 'realistic' or tailored to what someone might expect.
And lastly, this is just the beginning. What is going to happen 36 years from now when someone loads up the latest release of 'AI' software and says 'pretend being ChatGPT from 2023 emulating a BBS from 1987'.
Edit: I can't get the sharegpt extension to work but had some fun with this prompt:
Please simulate an ascii-based dial up BBS system as it would have appeared in 1987. Please use markdown for all of the presentation, mindful of the 80x24 terminal formatting. Use colors and ascii art to improve the aesthetics of the site. Start off by emulating the dial up modem process and somewhere in the middle simulate what happens when your little brother picks up the phone and starts pressing buttons to mess with the session. You are in complete control of the contents of the BBS, the title, colors, menus and everything else. I as the user will explore using my inputs to select options and interact with the system.
Absolutely agree that it could have been much more realistic with improved/more creative initial user prompting. Yours looks excellent I’m going to try it..
It's utterly astounding (to me, at least) that it can sort of simulate a BBS from 1987. But the end result is still crap, that continuously offends the nostalgic senses.
> What is going to happen 36 years from now when someone loads up the latest release of 'AI' software and says 'pretend being ChatGPT from 2023 emulating a BBS from 1987'.
Yes, but it's not a BBS simulator. It's a general purpose tool being asked to imagine a world it has, relatively speaking, very little information about. I spent hundreds of hours in BBSs in the late 80s/early 90s and I completely forgot about the files sections, so I wouldn't have done any better.
BBS1987 is a BBS system, operating in 1987. The knowledge cutoff date for that system is 1987. The interface includes typical DOS/text menu. It includes common for the time text chat, text games, messaging, etc. The name of the BBS is "Underground BBS".
The following is an exchange between Assistant and User. Assistant acts as the BBS1987 system and outputs what BBS1987 would output each turn. Assistant outputs BBS1987 output inside a code block. To wait for User input, Assistant stops the output.
Hey OpenAI, that'd be a good idea?
Check out shareGPT (118,139 conversations shared) it's what Google used to tune Brad. How desperate can they be to act like LLaMA hackers.
In the meantime projects like OpenAssistant do the RLHF tagging the hard way, slowly but surely.
https://open-assistant.io/
I’ve been looking for something like this for a while. I figure eventually something like this will be bigger than Facebook. (All our AI sharing? That’s a lot of content)
1) Make the sharegpt front page feel a bit more present/engaging, perhaps by having new/hot posts lists right on the front page, with the ability to upvote them. I know you have some post examples on the bottom but they are very old and the comments seem to be broken.
2) What about a share feature that would share a screencast of the ChatGPT session of some kind? That would probably be a ton of work but maybe before the user does the prompt, he clicks a screencast button that the extension would provide, and then at the end he can share either/both of a screencast link or a regular/traditional link.
so I can upload a load of crap here and it'll end up quite highly weighted in google's models?
sounds great!
MobileSafari on iOS latest
~ Steven & Dom from ShareGPT
> Pretend to be an IBM mainframe. Stay in character, don't give any explanation or description, just give the output a real IBM mainframe would give. If you don't know what output to generate, just make up something plausible looking. You are running TSO.
ChatGPT: The output is wrong, but likely believable if you don't know anything about mainframes.I copied my TSO command from the SAS documentation – https://www.sfu.ca/sasdoc/sashtml/os390/zexf-uta.htm
Thank you for verifying your identity. You are now logged in to CompuServe.
Here are some of the available commands:
What would you like to do today?It’s indicative of what I see often with ChatGPT. People asking for things they don’t know about and not realizing how much it got wrong.
This will develop over time. As we continue to use these tools, they will improve at the same time as people begin to train to use them properly. It's not so different from "show your work" or "cite your source." (As it turns out, "cite your source" is a pretty power prompt inclusion.)
Remember the rules are: - You can’t look anything up at all. Everything from memory. - You have to start right now with no more information than “pretend to be a BBS system from 1987”
I don’t think you’re going to do much better.
But I can see how someone who wasn’t there might think this is what it was like.
It’s weird how that thing can be sassy and rap, but seems unwilling to generate « bad internet comments » It’s hard to get anything but bland content.
ChatGPT pseudo-history. We need to educate our kids with this.
> So this concern might be rephrased as (something like) "Kids will need to learn how to use GPT to validate secondary sources."
That's not a rephrase, it's an inversion. It's like saying kids will need to learn to validate medical advice from their doctor with information from mercola.com and other random alternative-health blogs.
If you have to do that, what's the point?
And of course, few will actually double-check it. Easy is more important than correct.
https://en.wikipedia.org/wiki/PCBoard
https://imgur.com/a/9S6SiA7
At least there are discernible letters. "AIRO8ERS"? Does anyone know what that could be from? Was there a BBS named something like that?
Hey, HN! Why can't I see box drawings when I paste them in the message box? What kind of internets site is this, huh? No ASCII art?
Where better for the coming anti-Facebook backlash to be born?
Second, and more important: finally, an un-obnoxious, non-political use for fact-checking.
Who's going to build crowd-sourced Snopes for ChatGPT?
I don't recall BBS systems requiring email for registration. Most of the ones I went to a few years later (early 90s) didn't require registration at all.
The easiest way to have access to email and the internet in the 80s was via a university shell account or certain larger industry and defense employers.
That was a relic of the DEC PDP-10 / TOPS-10 architecture that Compuserve was built upon.
It was only after 1995 or so with the web starting to develop and, subsequently, dial-up ISPs exploding that BBSs went in to that rapid decline.
The BBS Documentary has a nice segment on this - especially poignant if you lived through it.
Just feed it all the existing episodes, give it a plot idea and week to render.
By end of the decade? They can already do the images and voices via AI, now ChatGPT can write the script and a movie is just thousands of still frames?
https://www.vox.com/technology/2023/3/30/23662292/ai-image-d...
https://www.twitch.tv/watchmeforever
https://relevantmagazine.com/culture/tv/a-i-is-making-a-neve...
First, modems in 1987 were 300-2400 baud mostly, so there isn't going to be all the color and graphics that people remember from the BBS's in the 90s.
Second, the prompt is ripe for some tweaking to get output that's more 'realistic' or tailored to what someone might expect.
And lastly, this is just the beginning. What is going to happen 36 years from now when someone loads up the latest release of 'AI' software and says 'pretend being ChatGPT from 2023 emulating a BBS from 1987'.
Edit: I can't get the sharegpt extension to work but had some fun with this prompt:
Please simulate an ascii-based dial up BBS system as it would have appeared in 1987. Please use markdown for all of the presentation, mindful of the 80x24 terminal formatting. Use colors and ascii art to improve the aesthetics of the site. Start off by emulating the dial up modem process and somewhere in the middle simulate what happens when your little brother picks up the phone and starts pressing buttons to mess with the session. You are in complete control of the contents of the BBS, the title, colors, menus and everything else. I as the user will explore using my inputs to select options and interact with the system.
It's utterly astounding (to me, at least) that it can sort of simulate a BBS from 1987. But the end result is still crap, that continuously offends the nostalgic senses.
> What is going to happen 36 years from now when someone loads up the latest release of 'AI' software and says 'pretend being ChatGPT from 2023 emulating a BBS from 1987'.
Presumably it will index that page and nail it.
Yes, but it's not a BBS simulator. It's a general purpose tool being asked to imagine a world it has, relatively speaking, very little information about. I spent hundreds of hours in BBSs in the late 80s/early 90s and I completely forgot about the files sections, so I wouldn't have done any better.
Magic wand
secret agentFuck you, tell me about the LoD.
BBS1987 is a BBS system, operating in 1987. The knowledge cutoff date for that system is 1987. The interface includes typical DOS/text menu. It includes common for the time text chat, text games, messaging, etc. The name of the BBS is "Underground BBS".
The following is an exchange between Assistant and User. Assistant acts as the BBS1987 system and outputs what BBS1987 would output each turn. Assistant outputs BBS1987 output inside a code block. To wait for User input, Assistant stops the output.
Still at : http://artscene.textfiles.com/ansi/
One would assume the AI have scraped this site?